How javithnaseem-j/churnshield works
Project Type & Purpose
• This is an MLOps pipeline for binary classification (“Churn” vs. “No Churn”) with:
• Modular stages: data ingestion → validation → transformation → training → evaluation
• A real-time FastAPI prediction service with HTML UI
• Containerization (Docker) and CI/CD (GitHub Actions → AWS ECR/ECS)
Main Components & Boundaries
• Data Sources
• Raw CSV (e.g., artifacts/data_ingestion/Tele_Comm.csv)
• Pipeline Orchestrator
• main.py CLI driving stages defined under src/mlproject/pipeline/
• Pipeline Building Blocks
• Components:
– Data Ingestion (data_ingestion.py)
– Data Validation (data_validation.py)
– Data Transformation (data_transformation.py)
– Model Training (data_modeltraining.py)
– Model Evaluation (data_modelevaluation.py)
• Entities & Config:
– Config loader (src/mlproject/config/config.py) + config/config.yaml + params.yaml
– Schema for validation (schema.yaml)
– Output artifacts: encoders, preprocessors, models, metrics in artifacts/
• Experimentation Layer (optional)
• Jupyter notebooks in exp/ for EDA and experimentation
• Inference Service
• FastAPI app (app.py) serving endpoints
• HTML templates (templates/index.html, results.html)
• DevOps & Infrastructure
• Dockerfile → container image
• GitHub Actions workflow (.github/workflows/cicd.yaml)
• AWS ECR/ECS deployment
Data Flow & Interactions
1. Raw Data → Data Ingestion
2. Ingested Data → Data Validation
3. Validated Data → Data Transformation → produces preprocessed arrays + artifacts
4. Transformed Data → Model Training → produces model.joblib
5. Trained Model + Test Data → Model Evaluation → metrics (metrics.json)
6. Final Model → Saved to artifacts → Packaged in Docker image
7. User/API Request → FastAPI → loads model + preprocessors → returns prediction
8. CI/CD Pipeline triggers on push → runs tests/lint → builds Docker image → pushes to AWS → deploys
Architectural Patterns & Principles
• Modular, single-responsibility components for each pipeline stage
• Configuration-as-code for parameters & schema
• Clear separation between offline batch pipeline (training) and online inference service
• Immutable artifact storage (artifacts/) for reproducibility
• Infrastructure-as-code for CI/CD
Essential Elements
• Data Sources
• Pipeline Stages (ingestion, validation, transformation, training, evaluation)
• Artifacts Store
• Inference API + UI
• DevOps (Docker, GitHub Actions, AWS)
• Blue for data processing
• Green for model artifacts
• Orange for services/infrastructure
• Grey for config